Concepts
These pages explain the agent-specific technologies and architecture boundaries that are easy to confuse when first entering the Agentic Web ecosystem.
Each concept page includes a short use case and a small example message, metadata document, or tool definition.
Core Concepts
- Agent Card: machine-readable metadata that describes an agent, its capabilities, skills, endpoints, protocols, and authentication requirements.
- Agent2Agent: protocol layer for communication and collaboration between independent agents.
- Model Context Protocol: protocol layer for connecting an agent or AI application to tools, resources, and reusable prompts.
- Agent Discovery: how clients find agents and evaluate discovery metadata.
- Identity Delegation: scoped access when agents act for users, services, or organizations.
- Trust Model: trust levels and evidence required before using an agent.
- Agent Registry: governed catalogs for approved agents and trust metadata.
- Tool Use: deterministic operations exposed to agents through protocols or APIs.
- Agent Memory: persistent and retrievable context beyond a single model call.
- Evaluation: scenario checks for behavior, security, reliability, and observability.
- Versioning and Governance: safe evolution of metadata, APIs, tools, and policies.
Comparisons
- MCP vs A2A: agent-to-tool and agent-to-agent boundaries.
- Agent Card vs OpenAPI: agent metadata and HTTP API contracts.
- RAG vs Memory: retrieval pattern and persistent context layer.
- Framework vs Protocol: implementation productivity and interoperability boundaries.
Related Reference Pages
- Technology Origins: historical context for when key technologies emerged, who developed them, and who stewards them.
- Specifications: machine-readable examples linked from concept and topic pages.
Quick Distinctions
| Concept | Main question | Typical artifact |
|---|---|---|
| Agent Card | What agent is this, and what can it do? | JSON document at /.well-known/agent-card.json |
| A2A | How does one agent talk to another agent? | Agent Card, task messages, artifacts, streaming or push updates |
| MCP | How does an agent use external tools and context? | MCP server exposing tools, resources, and prompts |
| OpenAPI | How does software describe HTTP APIs? | openapi.json or openapi.yaml |
| OAuth2/OIDC | How does access and identity work? | Access tokens, ID tokens, scopes, claims, discovery metadata |
| Agent Registry | Which agents are approved for this environment? | Registry entry with domain, version, trust level, protocols, and policy |